Types of medication errors are defined, error detection techniques are described, and the validity of several medication error studies is evaluated. A medication error is generally defined as a deviation from the physician's medication order as written on the patient's chart. In hospitals, medication errors occur at a rate of about one per patient per day. A dispensing error is one made by pharmacy staff when distributing medications to nursing units or directly to patients in an ambulatory-care pharmacy; the error rates for doses dispensed via the cart-filling process range from 0.87% to 2.9%. Categories of medication errors should be operationally defined before an investigation, and any allowable deviations from the physician's order should be clearly stated. Fourteen error category definitions are presented. Methods for detecting medication errors include anonymous self-reports (questionnaires), incident reports, the critical-incident technique (analyses of a large number of individual errors to identify common causal factors), and direct observation (including the disguised-observation and participant observer techniques). Observation is the best error detection method in terms of accuracy. Results of medication error studies were examined for validity and classified into one of four categories: (A) results should be accepted as reported, (B) results overestimate or underestimate the truth by a known amount, (C) results overestimate the truth by an unknown amount, and (D) results should not be accepted. All studies examined for validity used observation as the error detection technique. The following guidelines for observation-based medication error studies were established: The observer should follow the subject to the patient's bedside, the observer should witness patient consumption of each dose, the observer should not be familiar with patient drug regimens before observation, operational definitions must be used, and having an error validation committee can be advantageous. Future studies are needed that focus on the identification and testing of new error prevention methods that use the techniques described.